The Standing Wave
The hypotheses

The set we hold open

Fourteen durable questions about how the build-out resolves — read against the evidence every week.

The set

We organise every hypothesis across four layers of the build-out, from the capital and compute at the base to where it meets the market. The layers are the spine — a reader following one thread, the economics of compute, say, can trace it across every issue. These are the questions we hold open, grouped by layer. Each carries a dated horizon and a countable trigger — not a vibe — and each touches at least two others through shared dependencies, so the set is tracked as one system rather than a list of separate bets.

Current read — as of Signal № 004 (9 Jun 2026). Updated every Tuesday in the Signal.

Compute & Capex

The money and the silicon: data-centre spend and the cost of compute.

H1
Hyperscaler AI capex grows year-on-year through 2026, above prior consensus.
Strengthened
H2
A major hyperscaler signals capex digestion or pullback before end of 2026.
Weakened · watching
H14
AI infrastructure unit economics show visible strain in the serious press by end of 2026.
Strengthened
H15
Vertical integration of power, data centres and frontier compute becomes structurally consequential by end of 2027.
Both ways
Physical Layer

Power, land, the grid, and even space: where AI pushes physical limits.

H3
Power, not chips, becomes the binding constraint on the build-out by end of 2027.
Strengthened
H7
US grid interconnect queues stay a binding constraint — median waits above three years — through 2027.
Strengthened
H8
Three or more US nuclear restart or small-modular-reactor commitments financially close by end of 2026.
Unchanged
H9
Gas turbine lead times remain a near-term power constraint through 2026.
Strengthened
Frontier Capability

How fast models advance, and the economics that separate the leaders.

H4
A frontier lab ships a credibly autonomous AI researcher by end of 2026.
Unchanged
H5
Inference cost per token keeps roughly halving every 9–12 months — fastest at the commodity end, flatter at the frontier.
Both ways
H6
China closes the frontier-model gap to under six months on cost and benchmarks by end of 2026, while paradigm-shift capability may widen.
Strengthened
Application & Deployment

Where AI lands: enterprise adoption, consumer integration, and a new wave of lean software.

H10
Robotaxi reaches unsupervised commercial deployment at scale — six figures of paid rides a week across multiple metros — by end of 2027.
Both ways
H11
Humanoid robotics crosses a real deployment threshold — over 10,000 commercially deployed units — by end of 2026.
Unchanged
H12
Application-layer AI revenue exceeds $50B globally in 2026.
Strengthened

Fourteen hypotheses are active across the four layers; the set runs to H15, with H13 never activated. Resolved hypotheses are retired plainly in the quarterly review. The wording on this page is canonical — Signals carry weekly status reads only; any earlier shorthand in archived issues defers here.


New here? Why hypotheses? → explains the discipline behind the set.

Read the latest Signal

The set in motion — this week's read across the four layers.

Start with this week's Signal →